This document explores the skill of ECMWF’s seasonal forecast, to predict dry spells. ECMWF releases a forecast each month. This forecast includes projected total precipitation per month for 1 to 6 months ahead. The 1 month ahead is the month the forecast was released, and the release date is always the 13th of the month. I.e. the 1 month leadtime only becomes available when we are alread two weeks into that month.
ECMWF’s forecast is a probabilistic forecast, meaning it consists of several members (=models) each having their projected precipitation. This is what in this document is referred to as % of members, and can be interpreted as a probability of the event occurring.
This analysis
Set percentage and thereafter determine threshold
The figure below shows the distribution of probabilities with and without a dry spell per leadtime. From this figure it can be seen that the two distributions are not very separable.
Based on the misses and false alarms, the threshold is set to 220
Since we focus on leadtime 2 and 4, we only show these confusion matrices
Based on the misses and false alarms, the threshold is set to 240
Since we focus on leadtime 2 and 4, we only show these confusion matrices
Set threshold and thereafter determine percentage. In this case the percentage was set per lead time, where the percentage was chosen such that the number of months the forecast would trigger was closest to the number of months the observed precipitation was below the threshold. ### What is the percentage of ensemble members forecasted below a certain threshold? The figure below shows for how many months the % of ensemble members was below 170 mm. The 170 mm was the threshold set based on overlap of dry spells and monthly precipitation analysis.
We can see that the number of months lowers as the percentage increases, which is expected. While there are slight differences between lead times, the pattern is comparable.
To set the threshold of % of members, we look at the observed data and count how many months had <=170mm. This were 18 months. We then choose the % of members for each leadtime which has closest to 18 months forecasted <= 170mm
After setting the % of members, we get a list of forecasted months for which we would have triggered. Now we can compare these with the observed months with less than 170 mm of precipitation.
NOTE: false alarms is FP/(FP+TP)! i.e. the percentage of times the trigger was met but the monthly precipitation was not below 170
NOTE: false alarms is FP/(FP+TP)! i.e. the percentage of times the trigger was met but there was no dry spell